Generative artificial intelligence (GenAI) holds transformative potential for education, yet it is often introduced with a technology centered mindset that focuses on what the tools can do rather than how they support or hinder learning processes. Such technocentric approaches prioritize access and novelty over pedagogy, expecting learners to adapt to the technology instead of the other way around. This chapter promotes a human-centered alternative that begins with an understanding of how students learn and draws on educational and psychological theory to design GenAI tools that are grounded in established learning principles, support higher order thinking, and strengthen active engagement in learning processes. We review studies adopting this perspective and examine how the instructional possibilities of GenAI can be realized through theory-driven feature integration that addresses the cognitive, motivational, affective, and metacognitive processes involved in learning. This chapter first identifies the most frequently applied learning theories, including self-regulated learning, sociocultural theory, and cognitive load theory, and illustrates how they have been used to inform GenAI design. It then organizes the reviewed studies into three genres. Value-added research examines how theory informed GenAI features improve learning processes. Media comparison research evaluates GenAI in relation to traditional methods, revealing inversion, substitution, augmentation, and redefinition effects. Consequences research investigates how learners interact with GenAI and how these experiences shape cognition, motivation, affect, and self-regulation. Finally, we discuss the current limitations of GenAI educational research and propose directions for advancing a more empirically grounded and human-centered integration of GenAI in education.